Researchers Warn Leading AI Image Generators Vulnerable to Manipulation for Misinformation
-
Leading AI image generators can be manipulated to create misleading election-related images, according to a report by the Center for Countering Digital Hate.
-
Researchers tested Midjourney, DreamStudio, ChatGPT Plus, and Microsoft Image Creator and found they could produce realistic but misleading images about candidates or voting when prompted.
-
The AI platforms created misleading images in 41% of test runs, with Midjourney being most likely to produce questionable results.
-
ChatGPT Plus and Microsoft's tool blocked all candidate images, but did generate some concerning voting-related images.
-
The report urges AI companies to prevent "jailbreaking" of safety measures and social media platforms to better detect spread of misleading AI content.